Re: [PATCH v3 06/16] asprintf_chk: Ensure compatibility for both s390x and ppc64le

[Frederic Berat <fberat@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 3591 bytes --]

On Fri, Jun 30, 2023 at 7:51 PM Paul E Murphy <murphyp@linux.ibm.com> wrote:

>
>
> On 6/28/23 3:42 AM, Frédéric Bérat via Libc-alpha wrote:
> > If ldbl_* macros are used for asprintf, ABI gets broken on s390x,
> > if it isn't, ppc64le isn't building due to multiple asm redirections.
> >
> > This is due to the inclusion of bits/stdio-lbdl.h for ppc64le whereas it
> > isn't for s390x. This header creates redirections, which are not
> > compatible with the ones generated using libc_hidden_builtin_def.
> > Yet, we can't use libc_hidden_ldbl_proto on s390x since it will not
> > create a simple strong alias (e.g. as done on x86_64), but a versioned
> > alias, leading to ABI breakage.
> > ---
> >   debug/asprintf_chk.c | 12 ++++++++++++
> >   include/stdio.h      |  5 +++++
> >   2 files changed, 17 insertions(+)
> >
> > diff --git a/debug/asprintf_chk.c b/debug/asprintf_chk.c
> > index 06d112de00..7da1cddacb 100644
> > --- a/debug/asprintf_chk.c
> > +++ b/debug/asprintf_chk.c
> > @@ -36,5 +36,17 @@ ___asprintf_chk (char **result_ptr, int flag, const
> char *format, ...)
> >
> >     return ret;
> >   }
> > +#if defined __LDBL_COMPAT || __LDOUBLE_REDIRECTS_TO_FLOAT128_ABI == 1 >
> +/* This is needed since <bits/stdio-lbdl.h> is included in this
> case, leading to
> > + * multiple asm redirection of the same symbol
> > + */
> >   ldbl_hidden_def (___asprintf_chk, __asprintf_chk)
> >   ldbl_strong_alias (___asprintf_chk, __asprintf_chk)
> > +#else
> > +/* Yet, we can't use ldbl_* macros on some systems, even if they don't
> fall in
> > + * the first case as this leads to ABI breakage due to the
> long_double_symbol
> > + * aliasing, which is versionned.
>
> s/versionned/versioned/
>
> It has been a few years since I really looked at this macros, why only
> asprintf and not the other printf functions?  I find the ABI comment
> hard to follow, could it include more details about how it would be
> broken by using the ldbl_* macros?
>

I didn't do this for other functions as it didn't appear to be necessary.
Yes, I can be more specific about the breakage. Basically, the
__asprintf_chk routine disappears from the ABI list on s390x, which breaks
linkage. That is due to the fact that ldbl_strong_alias may create
*versioned* alias (depending on build options), while strong_alias never
does. I'll expand on the comment, I can reproduce the error to be explicit
about it.


>
> Also, does patchset 5 in this series break ppc64le or s390x build
> without this patch?  This seems like it should be squashed with the
> previous patch.
>

On s390x yes, not ppc. I was afraid to kind of hide this change in the
middle of the other changes, that's why I kept them separated, so that this
gets carefully reviewed.
I can squash this into patch 5 if you are fine with it.


> > + */
> > +strong_alias (___asprintf_chk, __asprintf_chk)
> > +libc_hidden_builtin_def (__asprintf_chk)
> > +#endif
> > diff --git a/include/stdio.h b/include/stdio.h
> > index 3afb0ff77b..6755877911 100644
> > --- a/include/stdio.h
> > +++ b/include/stdio.h
> > @@ -279,7 +279,12 @@ rtld_hidden_proto (__libc_fatal)
> >
> >   libc_hidden_proto (__fgets_unlocked_chk)
> >
> > +#if defined __LDBL_COMPAT || __LDOUBLE_REDIRECTS_TO_FLOAT128_ABI == 1
> >   libc_hidden_ldbl_proto (__asprintf_chk)
> > +#else
> > +libc_hidden_proto (__asprintf_chk)
> > +#endif
> > +
> >   libc_hidden_ldbl_proto (__fprintf_chk)
> >   libc_hidden_ldbl_proto (__sprintf_chk)
> >   libc_hidden_ldbl_proto (__vsprintf_chk)
>
>